CN112330706A - Mine personnel safety helmet segmentation method and device - Google Patents

Mine personnel safety helmet segmentation method and device Download PDF

Info

Publication number
CN112330706A
CN112330706A CN202011234672.0A CN202011234672A CN112330706A CN 112330706 A CN112330706 A CN 112330706A CN 202011234672 A CN202011234672 A CN 202011234672A CN 112330706 A CN112330706 A CN 112330706A
Authority
CN
China
Prior art keywords
target
superpixel
block
pixel
super
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN202011234672.0A
Other languages
Chinese (zh)
Inventor
任安祥
李晓宇
李萍
席庆荣
田柏林
王怀群
陈耕
王文清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Coal Mining Electric Equipment Technical Development Co ltd
Beijing University of Technology
Original Assignee
Beijing Coal Mining Electric Equipment Technical Development Co ltd
Beijing University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Coal Mining Electric Equipment Technical Development Co ltd, Beijing University of Technology filed Critical Beijing Coal Mining Electric Equipment Technical Development Co ltd
Priority to CN202011234672.0A priority Critical patent/CN112330706A/en
Publication of CN112330706A publication Critical patent/CN112330706A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

The application relates to a mine personnel safety helmet segmentation method and a mine personnel safety helmet segmentation device, wherein the method comprises the following steps: extracting color feature vectors, texture feature vectors and target contour features of a plurality of superpixel blocks formed by target image granulation; inputting the color feature vector and the texture feature vector into a pre-trained Support Vector Machine (SVM) so as to divide all superpixel blocks into a target superpixel block and a background superpixel block; correcting the misclassified superpixel blocks according to the target contour characteristics; judging whether a background pixel point exists in the corrected target superpixel block or not; and if not, segmenting the target image according to the target super pixel block. The method and the device can reduce the difficulty of dividing the target of the person, thereby being beneficial to the application of the technologies such as person detection, identification, positioning and tracking and the like.

Description

Mine personnel safety helmet segmentation method and device
Technical Field
The application relates to the technical field of image segmentation, in particular to a mine personnel safety helmet segmentation method and device.
Background
Mine personnel safety helmet segmentation is to independently separate safety helmet pixel areas in personnel images by using a method. The safety helmet segmentation is one of key technologies for realizing intelligent video monitoring of coal mine personnel, is a core content of computer vision in mine intelligent monitoring application, can promote application of related technologies such as mine personnel scheduling management, target detection and identification and position information prediction thereof based on the computer vision, and can improve the management and control efficiency of personnel operation areas.
There are many image segmentation methods, such as threshold method based on image gray feature, pixel point region growing method, edge detection method, graph segmentation method based on graph theory, deep learning neural network method, etc., but these methods have the following problems: the key of the threshold method lies in the reasonable selection of the gray threshold, which is suitable for processing the image with clear gray boundary between the target and the background; the region growing method has ideal image segmentation effect for lacking prior information, but is easy to cause excessive segmentation; the edge detection method is easy to have discontinuous boundary contour lines and poor in image region structure; the graph segmentation method requires a user to specify a target and a background in the image segmentation process, and is not suitable for automatic segmentation; the deep neural network segmentation method requires a large amount of input data, a long processing time and high requirements on computer hardware. The results in mine video image processing are difficult to meet with practical requirements.
Disclosure of Invention
In order to solve the problems in the background art, the application provides a mine personnel safety helmet segmentation method and a mine personnel safety helmet segmentation device.
In a first aspect, the present application provides a mine personnel safety helmet segmentation method, including:
extracting color feature vectors, texture feature vectors and target contour features of a plurality of superpixel blocks formed by target image granulation;
inputting the color feature vector and the texture feature vector into a pre-trained Support Vector Machine (SVM) so as to divide all superpixel blocks into a target superpixel block and a background superpixel block;
correcting the misclassified superpixel blocks according to the target contour features;
judging whether a background pixel point exists in the corrected target superpixel block or not;
and if not, segmenting the target image according to the target super pixel block.
Preferably, the determining whether a background pixel point exists in the corrected target superpixel block includes:
extracting a boundary mask of a superpixel area of the corrected target superpixel block;
by using
Figure 78274DEST_PATH_IMAGE001
An operator extracts the contour edge of the corrected target superpixel block;
calculating a difference set of a boundary mask of a superpixel region of the modified target superpixel block and a contour edge of the modified target superpixel block
Figure 729835DEST_PATH_IMAGE002
；
If it is
Figure 306310DEST_PATH_IMAGE003
Determining that no background pixel point exists in the corrected target superpixel block;
if it is
Figure 435940DEST_PATH_IMAGE004
Then determining saidAnd background pixel points exist in the corrected target superpixel blocks.
Preferably, the method further comprises:
if yes, classifying the corrected target superpixel blocks again to obtain target pixel point superpixel blocks and background pixel point superpixel blocks;
filtering the background pixel super-pixel block;
and segmenting the target image according to the super pixel block of the target pixel point.
Preferably, the classifying the corrected target superpixel block again includes:
according to the difference set
Figure 47050DEST_PATH_IMAGE002
And the boundary line of the pixel point of the super pixel block decomposes the super pixel block into the super pixel block of the target pixel point and the super pixel block of the background pixel point.
Preferably, the target in the target image is a safety helmet, and the target contour feature includes:
the safety helmet comprises a safety helmet image, and is characterized in that a zero slope point, a slope catastrophe point and at most two bulges exist on an outer contour line of the safety helmet, and the absolute value of the slope of a curve segment positioned among the zero slope point, the slope catastrophe point and the bulges is monotonously changed in the x direction or the y direction in a plane where the safety helmet image is positioned.
Preferably, the correcting the misclassified superpixel blocks according to the target contour features comprises:
extracting a boundary mask of a superpixel region of the target superpixel block and analyzing the change characteristic of the slope of a straight line between pixels on the boundary mask;
and modifying the category of the target superpixel block according to the change characteristics and the target contour characteristics.
Preferably, the modifying the category to which the target super pixel block belongs according to the variation characteristic and the target contour feature includes:
judging whether the change characteristics accord with the target contour features or not;
if yes, reserving the target superpixel block;
if not, detecting the number of the superpixel blocks contained in the boundary mask of the superpixel region of the target superpixel block;
based on the number, modifying the class to which the target superpixel block belongs according to the change characteristic and the target contour feature.
Preferably, the extracting the color feature vectors and the texture feature vectors of the plurality of super-pixel blocks formed by the target image granulation includes:
selecting
Figure 869512DEST_PATH_IMAGE005
、
Figure 870966DEST_PATH_IMAGE006
、
Figure 866604DEST_PATH_IMAGE007
、
Figure 269904DEST_PATH_IMAGE008
Describing color features by four color models to obtain the color feature vector;
and describing the texture feature vector by adopting a multi-order matrix of the gray distribution mean value of the pixel values in the super pixel block, wherein the texture feature vector comprises four attributes of dispersion degree, variance, skewness and kurtosis.
Preferably, the color feature vector is calculated by using the following formula:
Figure 263268DEST_PATH_IMAGE009
in the formula (I), the compound is shown in the specification,
Figure 814335DEST_PATH_IMAGE010
in order to be a color feature vector,
Figure 285767DEST_PATH_IMAGE011
、
Figure 809153DEST_PATH_IMAGE012
、
Figure 35735DEST_PATH_IMAGE013
、
Figure 11781DEST_PATH_IMAGE014
、
Figure 286904DEST_PATH_IMAGE015
、
Figure 727113DEST_PATH_IMAGE016
、
Figure 796700DEST_PATH_IMAGE017
、
Figure 525622DEST_PATH_IMAGE018
、
Figure 401174DEST_PATH_IMAGE019
、
Figure 633572DEST_PATH_IMAGE020
、
Figure 201957DEST_PATH_IMAGE021
、
Figure 152595DEST_PATH_IMAGE022
feature components on the four color models;
the dispersity, the variance, the skewness and the kurtosis of the texture feature vector are respectively calculated by adopting the following formula:
Figure 769521DEST_PATH_IMAGE023
Figure 918743DEST_PATH_IMAGE024
Figure 861291DEST_PATH_IMAGE025
Figure 33646DEST_PATH_IMAGE026
in the formula (I), the compound is shown in the specification,
Figure 251001DEST_PATH_IMAGE027
as the degree of dispersion,
Figure 457992DEST_PATH_IMAGE028
is the variance of the received signal and the received signal,
Figure 368179DEST_PATH_IMAGE029
in order to obtain the degree of skewness,
Figure 293409DEST_PATH_IMAGE030
in order to be the kurtosis,
Figure 252138DEST_PATH_IMAGE031
in order to be a gray scale level,
Figure 110373DEST_PATH_IMAGE032
a gray histogram function corresponding to the super pixel block.
In a second aspect, the present application provides a mine personnel safety helmet segmenting device, comprising:
the extraction module is used for extracting color feature vectors, texture feature vectors and target contour features of a plurality of super pixel blocks formed by target image granulation;
the classification module is used for inputting the color feature vector and the texture feature vector into a pre-trained Support Vector Machine (SVM) so as to divide all superpixel blocks into two types of target superpixel blocks and background superpixel blocks;
the correction module is used for correcting the category of the target superpixel block according to the target contour feature;
the judging module is used for judging whether the corrected target superpixel block has background pixel points;
and the segmentation module is used for segmenting the target image according to the target super-pixel block when the background pixel point does not exist in the corrected target super-pixel block.
In the mine personnel safety helmet segmentation method and device provided by the embodiment of the application, the target image is segmented into the plurality of superpixel blocks based on the SLIC model, the plurality of superpixel blocks are classified based on the Support Vector Machine (SVM), the classified superpixel blocks are subjected to class correction, and finally the target image is segmented by adopting the corrected superpixel blocks, so that the defects that the collected personnel video image has uneven illumination, distorted color information, random shadow distribution, difficulty in distinguishing the target and background boundaries and the like due to the influence of various severe conditions such as dust interference, high noise, poor illumination conditions and the like can be avoided, the difficulty in segmenting the personnel target is reduced, and the application of the technologies such as personnel detection, identification, positioning and tracking and the like is facilitated.
Drawings
FIG. 1 shows a schematic flow diagram of a mine personnel helmet segmentation method of an embodiment of the present application;
fig. 2 shows a block schematic diagram of a mine personnel safety cap separation device of an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The underground coal mine environment is special, and is influenced by various severe conditions such as dust interference, large noise, poor illumination conditions and the like, so that the collected personnel video images have the defects of uneven illumination, distorted color information, random shadow distribution, difficulty in distinguishing the target from the background boundary and the like, the personnel target segmentation difficulty is directly increased, and the application of the technologies such as personnel detection, identification, positioning and tracking is not facilitated.
The safety helmet is important safety protection equipment which must be worn by mine workers, is a necessary condition for guaranteeing the safety operation of the workers, and represents the existence of the workers. The safety helmet segmentation can promote the application of related technologies such as mine personnel scheduling management, target detection and identification and position information prediction thereof based on computer vision, can improve the management and control efficiency of the personnel working area, and can also effectively reduce the complexity of the whole body of the segmented personnel and the data processing capacity of a compression algorithm to personnel images.
Therefore, the application provides a mine personnel safety helmet segmentation method and device. Before introducing the method and the device for segmenting the safety helmet of the mine personnel, firstly, a support vector machine is introduced
Figure 129144DEST_PATH_IMAGE033
The training process of (1).
During training, the collected images with the safety helmet can be divided into sample images and target images, and the sample images are used for training the support vector machine
Figure 541671DEST_PATH_IMAGE033
Support vector machine with training completed
Figure 366408DEST_PATH_IMAGE033
For segmenting the target image.
Firstly, carrying out pixel-level labeling on a safety helmet region in a sample image, and simultaneously recording the positions of pixel points of the safety helmet region in a labeled image
Figure 282411DEST_PATH_IMAGE034
Inputting different signals within a certain range
Figure 472084DEST_PATH_IMAGE035
Value of sample image
Figure 168644DEST_PATH_IMAGE036
Performing superpixel granulation, and extracting the pixel point position of each superpixel in the sample image
Figure 469176DEST_PATH_IMAGE037
And is and
Figure 770844DEST_PATH_IMAGE034
intersection finding operation
Figure 928156DEST_PATH_IMAGE038
Will be
Figure 49696DEST_PATH_IMAGE039
In (A) belong to
Figure 216235DEST_PATH_IMAGE034
All pixel points of the safety helmet super pixel block are used as the safety helmet super pixel block
Figure 106831DEST_PATH_IMAGE039
In the middle do not belong to
Figure 638306DEST_PATH_IMAGE034
All pixel points of (2) are used as background superpixel blocks.
Then, respectively extracting color characteristic vectors and texture characteristic vectors of the helmet super-pixel block and the background super-pixel block in the sample image, combining the color characteristic vectors and the texture characteristic vectors as characteristic variables, and adopting the characteristic variables to support a vector machine
Figure 43880DEST_PATH_IMAGE033
And (5) training.
In some embodiments, the actual characteristics of the image of the downhole personnel and the color characteristic requirements of the task of segmenting the safety helmet are combined to select
Figure 217372DEST_PATH_IMAGE005
、
Figure 962474DEST_PATH_IMAGE006
、
Figure 930430DEST_PATH_IMAGE007
、
Figure 823300DEST_PATH_IMAGE008
Four color models describe color characteristics. Therefore, the feature components of the superpixel of the target image under the above four color models can form a 12-dimensional color feature vector, which can be calculated by using the following formula:
Figure 269325DEST_PATH_IMAGE009
in the formula (I), the compound is shown in the specification,
Figure 931250DEST_PATH_IMAGE010
in order to be a color feature vector,
Figure 70107DEST_PATH_IMAGE011
、
Figure 387956DEST_PATH_IMAGE012
、
Figure 434410DEST_PATH_IMAGE013
、
Figure 419683DEST_PATH_IMAGE014
、
Figure 463863DEST_PATH_IMAGE015
、
Figure 331324DEST_PATH_IMAGE016
、
Figure 119152DEST_PATH_IMAGE017
、
Figure 958932DEST_PATH_IMAGE018
、
Figure 236329DEST_PATH_IMAGE019
、
Figure 263191DEST_PATH_IMAGE020
、
Figure 120289DEST_PATH_IMAGE021
、
Figure 611313DEST_PATH_IMAGE022
are feature components on the four color models.
In some embodiments, due to the fact that the pixel regions of the safety helmet of the underground personnel, the skin of the personnel, the work clothes, the environment background and the like have obvious texture differences, the method for distinguishing the super pixels of the safety helmet and the non-safety helmet by using the texture features is an effective segmentation method. The super-pixel histogram can reflect the frequency of the pixel gray value in the super-pixel of the target image on each gray level, and can select four attributes of dispersion degree, variance, skewness and kurtosis to represent a texture feature vector, which can be calculated by adopting the following formula:
Figure 997295DEST_PATH_IMAGE040
Figure 42611DEST_PATH_IMAGE041
Figure 234558DEST_PATH_IMAGE042
Figure 517772DEST_PATH_IMAGE026
in the formula (I), the compound is shown in the specification,
Figure 605814DEST_PATH_IMAGE027
as the degree of dispersion,
Figure 669585DEST_PATH_IMAGE028
is the variance of the received signal and the received signal,
Figure 602906DEST_PATH_IMAGE029
in order to obtain the degree of skewness,
Figure 6205DEST_PATH_IMAGE030
in order to be the kurtosis,
Figure 61886DEST_PATH_IMAGE031
in order to be a gray scale level,
Figure 550636DEST_PATH_IMAGE032
a gray histogram function corresponding to the super pixel block.
Then, the characteristic variables
Figure 84386DEST_PATH_IMAGE043
Is defined as:
Figure 607771DEST_PATH_IMAGE044
finally, setting a safety helmet superpixel block as a positive sample and a label as '1'; setting a background super-pixel block as a negative sample and a label as '0'; and adopts an automatic hyper-parameter optimization mode to train the support vector machine
Figure 772036DEST_PATH_IMAGE033
。
In some embodiments, cross-validation may be employed to evaluate
Figure 544820DEST_PATH_IMAGE033
The accuracy of the classification of the sample images can be calculated, in particular
Figure 85523DEST_PATH_IMAGE033
Is predicted by the error
Figure 463414DEST_PATH_IMAGE045
Evaluating the accuracy of its classification, of a certain value
Figure 595318DEST_PATH_IMAGE045
It is described that
Figure 324240DEST_PATH_IMAGE033
Probability of misprediction for unknown test samples. For example, to increase
Figure 137475DEST_PATH_IMAGE033
The prediction accuracy of the method is trained by adopting an automatic hyper-parameter optimization mode
Figure 432190DEST_PATH_IMAGE033
Then calculates again after being optimized by hyper-parameters
Figure 938258DEST_PATH_IMAGE033
At the time of mispredicting the loss value
Figure 888897DEST_PATH_IMAGE046
Illustrating training by hyper-parametric optimization
Figure 568140DEST_PATH_IMAGE033
The prediction error for an unknown test sample is relatively small.
The method for segmenting the safety helmet of the mine personnel is described in detail below.
Fig. 1 shows a schematic flow diagram of a mine personnel safety cap segmentation method of an embodiment of the present application. As shown in fig. 1, the mine personnel safety helmet segmentation method and device comprises the following steps:
step 110, extracting color feature vectors, texture feature vectors and target contour features of a plurality of superpixel blocks formed by target image granulation.
In this embodiment, the method for granulating the target image is the same as the method for granulating the sample image, and the color feature vectors and texture feature vectors of the plurality of superpixel blocks formed by extracting and granulating the target image are also the same as the processing method in the sample image, which is not repeated herein.
In some embodiments, the objects in the image of the objects are helmets, which are produced strictly according to national standards GB2811-2019, are the most salient pixel regions in the image of mine personnel, having fixed geometric outlines in addition to a specific color class.
In the embodiment, the morphological characteristics of the shell part of the safety helmet are mainly studied, the main body part of the safety helmet except the brim and the peak is in a similar hemispherical shape, the external contour is formed by smooth curves, and the contour line of the safety helmet is
Figure 655044DEST_PATH_IMAGE047
Must have a zero slope point
Figure 597592DEST_PATH_IMAGE048
And slope discontinuity
Figure 832265DEST_PATH_IMAGE049
Two types of points, one or two pointed "bumps" are present on the outline of the helmet in the frontal image taken at the central position of the helmet, so that a maximum of two "bumps" may be present, and are located at the zero slope point
Figure 987303DEST_PATH_IMAGE048
Slope discontinuity point
Figure 459872DEST_PATH_IMAGE049
And a curved section in the region of the "bulge
Figure 104480DEST_PATH_IMAGE050
Absolute value of slope of
Figure 764132DEST_PATH_IMAGE051
Monotonously changes in the x direction or the y direction in the plane of the helmet image.
Thus, the target profile features may include: the safety helmet comprises a safety helmet body, a safety helmet image, a slope image and a slope image, wherein a zero slope point, a slope catastrophe point and at most two bulges are arranged on an outer contour line of the safety helmet body, and the absolute value of the slope of a curve section positioned between the zero slope point, the slope catastrophe point and the bulges is monotonously changed in the x direction or the y direction in a plane where the safety helmet image is positioned.
Step 120, inputting the color feature vector and the texture feature vector into a pre-trained support vector machine
Figure 988440DEST_PATH_IMAGE033
In the method, all the superpixel blocks are divided into two types of target superpixel blocks and background superpixel blocks.
And step 130, modifying the category of the target superpixel block according to the target contour characteristics.
In the embodiment, the support vector machine is supported due to participation
Figure 112253DEST_PATH_IMAGE033
The trained sample images lack global representativeness in the whole data sample set, and simultaneously, a support vector machine obtained through hyper-parameter optimization training
Figure 131025DEST_PATH_IMAGE033
The prediction capability of the test sample is limited, so that a small number of negative samples are wrongly divided into positive samples, and the positive samples need to be corrected.
In some embodiments, the target superpixel block may be modified by:
step 1301, extracting the boundary mask of the superpixel region of the target superpixel block and analyzing the change characteristics of the slope of the straight line between the pixels on the boundary mask.
In the present embodiment, a boundary mask of a superpixel region of a target superpixel block is extracted
Figure 340289DEST_PATH_IMAGE052
And adopting morphological expansion operator to process, then selecting boundary mask of superpixel region of target superpixel block
Figure 102709DEST_PATH_IMAGE052
The contour line of any point above
Figure 18712DEST_PATH_IMAGE047
Starting point of (2)
Figure 270702DEST_PATH_IMAGE053
Calculating the slope of the straight line between adjacent unit pixels in sequence according to the clockwise direction or the counterclockwise direction
Figure 170525DEST_PATH_IMAGE054
Until it returns to the starting point
Figure 471056DEST_PATH_IMAGE053
And analyzing the slope of the straight line
Figure 303883DEST_PATH_IMAGE054
The change characteristic of (c).
Step 1302, modifying the category of the target superpixel block according to the change characteristics and the target contour characteristics.
First, the boundary mask of the super pixel area of the target super pixel block is judged
Figure 664457DEST_PATH_IMAGE052
Generated from several superpixel blocks.
For example, boundary mask of superpixel region of target superpixel block
Figure 51576DEST_PATH_IMAGE052
Generated from a single superpixel block. Slope of straight line
Figure 952536DEST_PATH_IMAGE054
Is completely in accordance with the target contour feature, the change characteristics are judged as a safety helmet area which is divided by a superpixel block
Figure 843132DEST_PATH_IMAGE055
(ii) a Slope of straight line
Figure 640187DEST_PATH_IMAGE054
If the change characteristic of (a) is partially in accordance with the target profile feature, the slope of the line that will not be in accordance with the target profile feature
Figure 45760DEST_PATH_IMAGE054
Defined as the slope of an abnormal straight line
Figure 953673DEST_PATH_IMAGE054
By a region adjacency graph
Figure 761092DEST_PATH_IMAGE056
Retrieving slope of line due to abnormality
Figure 729048DEST_PATH_IMAGE054
Neighbor superpixel block corresponding to interval
Figure 559601DEST_PATH_IMAGE057
Extracting its boundary mask and
Figure 802364DEST_PATH_IMAGE058
generating a new superpixel region mask after fusing
Figure 933131DEST_PATH_IMAGE059
Updating the slope of the line
Figure 806409DEST_PATH_IMAGE054
And analyzing the change characteristics of the super-pixels, if the proportion of the parts conforming to the target contour features is increased, retaining the newly-added super-pixels, otherwise, regarding the super-pixels as misclassification samples, and filtering the samples until the slope of the straight line conforming to the target contour features
Figure 186575DEST_PATH_IMAGE054
The ratio of (a) is maximized.
As another example, boundary mask of superpixel region of target superpixel block
Figure 436290DEST_PATH_IMAGE052
Generated from a plurality of superpixel blocks. Slope of straight line
Figure 155985DEST_PATH_IMAGE054
Is determined to be a helmet area divided by a plurality of superpixel blocks when the change characteristics of (a) completely accord with the target contour characteristics
Figure 262481DEST_PATH_IMAGE055
(ii) a Slope of straight line
Figure 333205DEST_PATH_IMAGE054
In which only part of the slope of the straight line
Figure 121032DEST_PATH_IMAGE054
Is in accordance with the target contour feature and is in accordance with the slope of the straight line of the target contour feature
Figure 757550DEST_PATH_IMAGE054
And slope of line not conforming to target profile feature
Figure 972631DEST_PATH_IMAGE054
Simultaneously distributed over each super-pixel block boundary, then
Figure 265072DEST_PATH_IMAGE060
Retrieving the slope of a line that does not match the target profile feature
Figure 918907DEST_PATH_IMAGE054
Interval corresponds to
Figure 347614DEST_PATH_IMAGE057
Extracting its boundary mask and
Figure 999176DEST_PATH_IMAGE058
generating a new superpixel region mask after fusing
Figure 841230DEST_PATH_IMAGE059
Updating the slope of the line
Figure 970860DEST_PATH_IMAGE054
And analyzing the change characteristics to conform to the slope of the straight line of the target contour characteristics
Figure 316390DEST_PATH_IMAGE054
The new addition is reserved when the occupied proportion is increasedSuperpixels, and the steps are executed in a loop until the slope of the straight line conforming to the target contour characteristics
Figure 404432DEST_PATH_IMAGE054
Maximizing the proportion; if the slope of the straight line of the target contour feature is met
Figure 405886DEST_PATH_IMAGE054
In which the slope of the line only partially conforms to the profile feature of the target
Figure 135945DEST_PATH_IMAGE054
Is in accordance with the target contour feature and is in accordance with the slope of the straight line of the target contour feature
Figure 804824DEST_PATH_IMAGE054
And slope of line not conforming to target profile feature
Figure 798187DEST_PATH_IMAGE054
Distributed on different super-pixel boundaries, is composed of
Figure 83675DEST_PATH_IMAGE056
Retrieving the slope of a line that does not match the target profile feature
Figure 820687DEST_PATH_IMAGE054
Interval corresponds to
Figure 344072DEST_PATH_IMAGE057
And removing the target contour features to update the slope of the straight line conforming to the target contour features
Figure 305075DEST_PATH_IMAGE054
And analyzing the change characteristics to conform to the slope of the straight line of the target contour characteristics
Figure 281121DEST_PATH_IMAGE054
If the proportion is increased, the correction is effective, and the cyclic execution is carried out until the linear slope of the target contour characteristic is met
Figure 821824DEST_PATH_IMAGE054
Maximizing the proportion;
slope of straight line
Figure 996453DEST_PATH_IMAGE054
If the target contour feature is completely not met, the method is characterized by
Figure 331620DEST_PATH_IMAGE056
Retrieving the slope of the line
Figure 794962DEST_PATH_IMAGE054
Corresponding super pixels are removed one by one and the slope of the straight line is updated
Figure 936094DEST_PATH_IMAGE054
Slope of straight line
Figure 168492DEST_PATH_IMAGE054
And judging the target contour characteristics which are not met with the target contour characteristics as a background area, and filtering all superpixel blocks contained in the background area.
Step 140, determining whether the modified target superpixel block has background pixel points.
In some embodiments, the following steps may be employed for the determination:
in step 1401, the boundary mask of the superpixel region of the corrected target superpixel block is extracted.
Step 1402, adopt
Figure 674559DEST_PATH_IMAGE001
And the operator extracts the contour edge of the corrected target superpixel block.
Step 1403, calculate the difference set of the boundary mask of the superpixel region of the modified target superpixel block and the contour edge of the modified target superpixel block
Figure 687515DEST_PATH_IMAGE002
。
In step 1404, if
Figure 304441DEST_PATH_IMAGE003
And determining that no background pixel point exists in the corrected target superpixel block.
Step 1405, if
Figure 453663DEST_PATH_IMAGE004
And determining that background pixel points exist in the corrected target superpixel block.
It should be noted that, if it is determined that the background pixel does not exist in the corrected target super pixel block, step 150 is executed, and if it is determined that the background pixel exists in the corrected target super pixel block, step 160 is executed.
And 150, segmenting the target image according to the target superpixel blocks.
And 160, classifying the corrected target superpixel blocks again to obtain target pixel point superpixel blocks and background pixel point superpixel blocks, filtering the background pixel point superpixel blocks, and segmenting the target image according to the target pixel point superpixel blocks.
In this embodiment, the revised target superpixel blocks are again classified according to the difference set
Figure 130632DEST_PATH_IMAGE002
The boundary line of the pixel point of the super pixel block decomposes the super pixel block into a target pixel super pixel block and a background pixel super pixel block.
In the embodiment of the application, a target image is granulated into a plurality of superpixel blocks based on a SLIC model, the superpixel blocks are classified based on a Support Vector Machine (SVM), the classified superpixel blocks are corrected, and finally the corrected superpixel blocks are adopted to divide the target image, so that the defects that the video image of an acquired person is uneven in illumination, distorted in color information, random in shadow distribution, difficult to distinguish between the target and a background boundary and the like due to the influence of various severe conditions such as dust interference, high noise, poor illumination condition and the like can be avoided, the difficulty in dividing the person target is reduced, and the application of the technologies such as person detection, identification, positioning and tracking and the like is facilitated.
It should be noted that, for simplicity of description, the above-mentioned method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present application is not limited by the order of acts described, as some steps may occur in other orders or concurrently depending on the application. Further, those skilled in the art should also appreciate that the embodiments described in the specification are exemplary embodiments and that the acts and modules referred to are not necessarily required in this application.
The above is a description of method embodiments, and the embodiments of the present application are further described below by way of apparatus embodiments.
Fig. 2 shows a block schematic diagram of a mine personnel safety cap separation device of an embodiment of the present application. As shown in fig. 2, the mine personnel safety helmet dividing device includes:
the extracting module 210 is configured to extract color feature vectors, texture feature vectors, and target contour features of a plurality of super-pixel blocks formed by the target image granulation.
The classification module 220 is configured to input the color feature vectors and the texture feature vectors into a pre-trained support vector machine SVM, so as to divide all superpixel blocks into two types, namely a target superpixel block and a background superpixel block.
And a correction module 230 for correcting the misclassified superpixel blocks according to the target contour features.
And the judging module 240 is configured to judge whether a background pixel point exists in the corrected target superpixel block.
And a segmentation module 250, configured to segment the target image according to the target super-pixel block when there is no background pixel point in the corrected target super-pixel block.
In some embodiments, the determining module 240 is specifically configured to:
extracting a boundary mask of a superpixel area of the corrected target superpixel block;
by using
Figure 302987DEST_PATH_IMAGE001
The operator extracts the contour edge of the corrected target superpixel block;
calculating a difference set between a boundary mask of a superpixel region of the modified target superpixel block and a contour edge of the modified target superpixel block
Figure 785921DEST_PATH_IMAGE002
；
If it is
Figure 992911DEST_PATH_IMAGE003
If so, determining that no background pixel point exists in the corrected target superpixel block;
if it is
Figure 840782DEST_PATH_IMAGE004
And determining that background pixel points exist in the corrected target superpixel block.
In some embodiments, the segmentation module 250 is further configured to classify the corrected target super-pixel block again when a background pixel exists in the corrected target super-pixel block, so as to obtain a target pixel super-pixel block and a background pixel super-pixel block; filtering background pixel super-pixel blocks; and segmenting the target image according to the superpixel blocks of the target pixel points.
In some embodiments, the segmentation module 250 is further configured to segment the difference set according to the difference set
Figure 562750DEST_PATH_IMAGE002
The boundary line of the pixel point of the super pixel block decomposes the super pixel block into a target pixel super pixel block and a background pixel super pixel block.
In some embodiments, the target in the target image is a hard hat, and the target contour features include: the safety helmet comprises a safety helmet body, a safety helmet image, a slope image and a slope image, wherein a zero slope point, a slope catastrophe point and at most two bulges are arranged on an outer contour line of the safety helmet body, and the absolute value of the slope of a curve section positioned between the zero slope point, the slope catastrophe point and the bulges is monotonously changed in the x direction or the y direction in a plane where the safety helmet image is positioned.
In some embodiments, the modification module 230 is specifically configured to:
extracting a boundary mask of a superpixel region of a target superpixel block and analyzing the change characteristic of the slope of a straight line between pixels on the boundary mask;
and correcting the misclassified superpixel blocks according to the change characteristics and the target contour characteristics.
In some embodiments, the modification module 230 is further specifically configured to:
judging whether the change characteristics accord with the target contour characteristics;
if yes, reserving the target superpixel block;
if not, detecting the number of the superpixel blocks contained in the boundary mask of the superpixel region of the target superpixel block;
and modifying the category to which the target superpixel block belongs according to the change characteristics and the target contour features on the basis of the number.
In some embodiments, the extraction module 210 is specifically configured to:
selecting
Figure 787058DEST_PATH_IMAGE005
、
Figure 848555DEST_PATH_IMAGE006
、
Figure 929643DEST_PATH_IMAGE007
、
Figure 76591DEST_PATH_IMAGE008
Describing color features by the four color models to obtain color feature vectors;
and describing texture feature vectors by adopting a multi-order matrix of the gray distribution mean value of pixel values in the super-pixel blocks, wherein the texture feature vectors comprise four attributes of dispersion degree, variance, skewness and kurtosis.
In some embodiments, the color feature vector is calculated using the following equation:
Figure 839010DEST_PATH_IMAGE009
in the formula (I), the compound is shown in the specification,
Figure 817331DEST_PATH_IMAGE010
in order to be a color feature vector,
Figure 272583DEST_PATH_IMAGE011
、
Figure 906827DEST_PATH_IMAGE012
、
Figure 269675DEST_PATH_IMAGE013
、
Figure 40185DEST_PATH_IMAGE014
、
Figure 666338DEST_PATH_IMAGE015
、
Figure 584616DEST_PATH_IMAGE016
、
Figure 688838DEST_PATH_IMAGE017
、
Figure 653469DEST_PATH_IMAGE018
、
Figure 184945DEST_PATH_IMAGE019
、
Figure 590518DEST_PATH_IMAGE020
、
Figure 498431DEST_PATH_IMAGE021
、
Figure 243533DEST_PATH_IMAGE022
feature components on the four color models;
the dispersity, the variance, the skewness and the kurtosis of the texture feature vector are respectively calculated by adopting the following formula:
Figure 8227DEST_PATH_IMAGE061
Figure 104359DEST_PATH_IMAGE062
Figure 550384DEST_PATH_IMAGE063
Figure 477889DEST_PATH_IMAGE064
in the formula (I), the compound is shown in the specification,
Figure 616746DEST_PATH_IMAGE027
as the degree of dispersion,
Figure 669015DEST_PATH_IMAGE028
is the variance of the received signal and the received signal,
Figure 981048DEST_PATH_IMAGE029
in order to obtain the degree of skewness,
Figure 966322DEST_PATH_IMAGE030
in order to be the kurtosis,
Figure 10501DEST_PATH_IMAGE031
in order to be a gray scale level,
Figure 612384DEST_PATH_IMAGE032
a gray histogram function corresponding to the super pixel block.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the described module may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
The above embodiments are preferred embodiments of the present application, and the protection scope of the present application is not limited by the above embodiments, so: all equivalent changes made according to the structure, shape and principle of the present application shall be covered by the protection scope of the present application.

Claims (10)

1. A mine personnel safety helmet segmentation method is characterized by comprising the following steps:
extracting color feature vectors, texture feature vectors and target contour features of a plurality of superpixel blocks formed by target image granulation;
inputting the color feature vector and the texture feature vector into a pre-trained Support Vector Machine (SVM) so as to divide all superpixel blocks into a target superpixel block and a background superpixel block;
correcting the misclassified superpixel blocks according to the target contour features;
judging whether a background pixel point exists in the corrected target superpixel block or not;
and if not, segmenting the target image according to the target super pixel block.
2. The method of claim 1, wherein said determining whether a background pixel point exists in the modified target superpixel block comprises:
extracting a boundary mask of a superpixel area of the corrected target superpixel block;
by using
Figure 137165DEST_PATH_IMAGE001
An operator extracts the contour edge of the corrected target superpixel block;
calculating a difference set of a boundary mask of a superpixel region of the modified target superpixel block and a contour edge of the modified target superpixel block
Figure 711366DEST_PATH_IMAGE002
；
If it is
Figure 192026DEST_PATH_IMAGE003
Determining that no background pixel point exists in the corrected target superpixel block;
if it is
Figure 546784DEST_PATH_IMAGE004
And determining that background pixel points exist in the corrected target superpixel block.
3. The method of claim 2, further comprising:
if yes, classifying the corrected target superpixel blocks again to obtain target pixel point superpixel blocks and background pixel point superpixel blocks;
filtering the background pixel super-pixel block;
and segmenting the target image according to the super pixel block of the target pixel point.
4. The method of claim 3, wherein said re-classifying the modified target superpixel block comprises:
according to the difference set
Figure 872723DEST_PATH_IMAGE002
And the boundary line of the pixel point of the super pixel block decomposes the super pixel block into the super pixel block of the target pixel point and the super pixel block of the background pixel point.
5. The method of claim 1, wherein the object in the object image is a helmet, the object contour feature comprising:
the safety helmet comprises a safety helmet image, and is characterized in that a zero slope point, a slope catastrophe point and at most two bulges exist on an outer contour line of the safety helmet, and the absolute value of the slope of a curve segment positioned among the zero slope point, the slope catastrophe point and the bulges is monotonously changed in the x direction or the y direction in a plane where the safety helmet image is positioned.
6. The method of claim 5, wherein said modifying the misclassified superpixel based on the target contour features comprises:
extracting a boundary mask of a superpixel region of the target superpixel block and analyzing the change characteristic of the slope of a straight line between pixels on the boundary mask;
and modifying the category of the target superpixel block according to the change characteristics and the target contour characteristics.
7. The method of claim 6, wherein said modifying the class to which the target superpixel block belongs according to the variance characteristic and the target contour characteristic comprises:
judging whether the change characteristics accord with the target contour features or not;
if yes, reserving the target superpixel block;
if not, detecting the number of the superpixel blocks contained in the boundary mask of the superpixel region of the target superpixel block;
based on the number, modifying the class to which the target superpixel block belongs according to the change characteristic and the target contour feature.
8. The method of claim 1, wherein the extracting color feature vectors and texture feature vectors of a plurality of superpixel blocks formed by target image granulation comprises:
selecting
Figure 629326DEST_PATH_IMAGE005
、
Figure 280887DEST_PATH_IMAGE006
、
Figure 795045DEST_PATH_IMAGE007
、
Figure 986992DEST_PATH_IMAGE008
Describing color features by four color models to obtain the color feature vector;
and describing the texture feature vector by adopting a multi-order matrix of the gray distribution mean value of the pixel values in the super pixel block, wherein the texture feature vector comprises four attributes of dispersion degree, variance, skewness and kurtosis.
9. The method of claim 8, wherein the color feature vector is calculated using the following equation:
Figure 801364DEST_PATH_IMAGE009
in the formula (I), the compound is shown in the specification,
Figure 358248DEST_PATH_IMAGE010
in order to be a color feature vector,
Figure 422019DEST_PATH_IMAGE011
、
Figure 355340DEST_PATH_IMAGE012
、
Figure 24218DEST_PATH_IMAGE013
、
Figure 814320DEST_PATH_IMAGE014
、
Figure 303070DEST_PATH_IMAGE015
、
Figure 305661DEST_PATH_IMAGE016
、
Figure 625784DEST_PATH_IMAGE017
、
Figure 524470DEST_PATH_IMAGE018
、
Figure 766095DEST_PATH_IMAGE019
、
Figure 103536DEST_PATH_IMAGE020
、
Figure 215848DEST_PATH_IMAGE021
、
Figure 613331DEST_PATH_IMAGE022
feature components on the four color models;
the dispersity, the variance, the skewness and the kurtosis of the texture feature vector are respectively calculated by adopting the following formula:
Figure 76674DEST_PATH_IMAGE023
Figure 155488DEST_PATH_IMAGE024
Figure 184624DEST_PATH_IMAGE025
Figure 956271DEST_PATH_IMAGE026
in the formula (I), the compound is shown in the specification,
Figure 641330DEST_PATH_IMAGE027
as the degree of dispersion,
Figure 320573DEST_PATH_IMAGE028
is the variance of the received signal and the received signal,
Figure 673057DEST_PATH_IMAGE029
in order to obtain the degree of skewness,
Figure 350026DEST_PATH_IMAGE030
in order to be the kurtosis,
Figure 584698DEST_PATH_IMAGE031
in order to be a gray scale level,
Figure 270895DEST_PATH_IMAGE032
a gray histogram function corresponding to the super pixel block.
10. A mine personnel safety helmet segmenting device, comprising:
the extraction module is used for extracting color feature vectors, texture feature vectors and target contour features of a plurality of super pixel blocks formed by target image granulation;
the classification module is used for inputting the color feature vector and the texture feature vector into a pre-trained Support Vector Machine (SVM) so as to divide all superpixel blocks into two types of target superpixel blocks and background superpixel blocks;
the correction module is used for correcting the category of the target superpixel block according to the target contour feature;
the judging module is used for judging whether the corrected target superpixel block has background pixel points;
and the segmentation module is used for segmenting the target image according to the target super-pixel block when the background pixel point does not exist in the corrected target super-pixel block.
CN202011234672.0A 2020-11-07 2020-11-07 Mine personnel safety helmet segmentation method and device Withdrawn CN112330706A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011234672.0A CN112330706A (en) 2020-11-07 2020-11-07 Mine personnel safety helmet segmentation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011234672.0A CN112330706A (en) 2020-11-07 2020-11-07 Mine personnel safety helmet segmentation method and device

Publications (1)

Publication Number Publication Date
CN112330706A true CN112330706A (en) 2021-02-05

Family

ID=74316463

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011234672.0A Withdrawn CN112330706A (en) 2020-11-07 2020-11-07 Mine personnel safety helmet segmentation method and device

Country Status (1)

Country Link
CN (1) CN112330706A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114120358A (en) * 2021-11-11 2022-03-01 国网江苏省电力有限公司技能培训中心 Super-pixel-guided deep learning-based identification method for head-worn safety helmet of person
CN115439474A (en) * 2022-11-07 2022-12-06 山东天意机械股份有限公司 Rapid positioning method for power equipment fault

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114120358A (en) * 2021-11-11 2022-03-01 国网江苏省电力有限公司技能培训中心 Super-pixel-guided deep learning-based identification method for head-worn safety helmet of person
CN114120358B (en) * 2021-11-11 2024-04-26 国网江苏省电力有限公司技能培训中心 Super-pixel-guided deep learning-based personnel head-mounted safety helmet recognition method
CN115439474A (en) * 2022-11-07 2022-12-06 山东天意机械股份有限公司 Rapid positioning method for power equipment fault

Similar Documents

Publication Publication Date Title
Zhao et al. Cloud shape classification system based on multi-channel cnn and improved fdm
CN109961049B (en) Cigarette brand identification method under complex scene
CN110992381B (en) Moving object background segmentation method based on improved Vibe+ algorithm
CN112288706A (en) Automatic chromosome karyotype analysis and abnormality detection method
CN109684959B (en) Video gesture recognition method and device based on skin color detection and deep learning
CN107230188B (en) Method for eliminating video motion shadow
CN111738271B (en) Method for identifying blocked fruits in natural environment
CN108268823A (en) Target recognition methods and device again
CN105513053A (en) Background modeling method for video analysis
JP6932402B2 (en) Multi-gesture fine division method for smart home scenes
CN112330706A (en) Mine personnel safety helmet segmentation method and device
Campos et al. Discrimination of abandoned and stolen object based on active contours
Cheng et al. Urban road extraction via graph cuts based probability propagation
CN114241542A (en) Face recognition method based on image stitching
CN112446417B (en) Spindle-shaped fruit image segmentation method and system based on multilayer superpixel segmentation
CN110544262A (en) cervical cell image segmentation method based on machine vision
CN113723314A (en) Sugarcane stem node identification method based on YOLOv3 algorithm
CN112307894A (en) Pedestrian age identification method based on wrinkle features and posture features in community monitoring scene
CN108564020B (en) Micro-gesture recognition method based on panoramic 3D image
CN116524410A (en) Deep learning fusion scene target detection method based on Gaussian mixture model
Dantas et al. A deterministic technique for identifying dicotyledons in images
CN111415350B (en) Colposcope image identification method for detecting cervical lesions
CN213241250U (en) Miner safety helmet detection system
CN110599518B (en) Target tracking method based on visual saliency and super-pixel segmentation and condition number blocking
Sujatha et al. An innovative moving object detection and tracking system by using modified region growing algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20210205